Lots of people like sharing what they're about to eat on Instagram, but who knew those casual food snaps would end up constituting a massive dietary database for scientists to pick over long after the meal has been enjoyed?
A team of researchers in the US has combed through some 3 million geo-tagged food posts on Instagram and come away with an interesting finding: those living in communities with limited access to grocery stores – known as 'food deserts' – could be including notably higher amounts of fat, cholesterol, and sugars in their diets.
"The USDA identifies food deserts based on the availability of fresh food," said computer scientist Munmun De Choudhury from the Georgia Institute of Technology (Georgia Tech). "Instagram literally gives us a picture of what people are actually eating in these communities, allowing us to study them in a new way."
While there's no proof that the foods being depicted on Instagram are actually eaten by those who upload the images, previous research has found a correlation between what people put on social media and what they actually eat. But it also found a strong cultural influence – rich people tend to publicise healthier foods – which is something that could also be at play in this Instagram-based study.
According to the Georgia Tech researchers, people living in food deserts are more likely to be photographing (and eating) foods such as pork, mayonnaise, and cookies. In contrast, those with better access to grocery stores and the nutritious food for sale within them enjoy (and upload) more things like bagels, kale, and hummus.
"Fruits and vegetables are the biggest difference," said De Choudhury. "Forty-eight percent of posts from people in non-food deserts mention them. It's only 33 percent in food deserts."
By region, this is what people in different places Instagram more of:
- In the southeast: bacon, potatoes, and grits (food deserts) vs. collard greens, oranges, and peaches (non-food deserts).
- In the midwest: hamburgers, hot dogs, and brisket vs. beans, spinach, and kale.
- In the west: pie, beef, and sausage vs. quinoa, apple, and crab.
- In the southwest: barbeque, pork, and burritos vs. tomatoes, asparagus, and bananas.
Using Instagram as their data set, the researchers scanned through millions of geo-tagged posts looking for designators from a list of almost 600 food-related tags, such as 'tofu', 'chocolate', 'beef', and 'oatmeal'.
Their findings, being presented this week at the ACM Conference on Computer-Supported Cooperative Work and Social Computing in San Francisco, suggest that food consumed in food deserts is 5 to 17 percent higher in fat, cholesterol, and sugars than what what's being eaten in non-food deserts.
To calculate the differences, De Choudhury and her colleagues used USDA-provided nutritional values for almost 9,000 different foods. In what came as a surprise to the team, despite the contrast in food eaten by the two group profiles (food deserts and non-food deserts), the overall energy consumed was similar.
"[C]ounter-intuitively, calorie intake of the food posted by people in food deserts is not significantly higher than that in other locations," the authors write, "however fat, cholesterol, and sugar intake in food, as indicated by Instagram content, is notably high."
The researchers acknowledge there are significant limitations with their way of gauging what people are eating – for example, food portions can't be measured, and there's no way of knowing whether the food in the photograph was actually eaten by uploaders, or what percentage of their diet it actually constitutes. Local culture can also have a big influence on the types of food a person chooses to tell their friends about.
But it's hoped that this system will help scientists study communities where people don't have adequate access to healthy foods.
"Our results bear implications in how longitudinal inferences of nutritional and food deprivation status of areas derived from social media may be useful in improved detection of food deserts and thereby helping reduce inequalities in health," the team writes.